🏆 Data Science I - ML Competition: Predict Academic Performance

Welcome to the Machine Learning Kaggle Competition!

🎯 Objective

Use the provided dataset to predict students' final grades based on various features (study time, failures, absences, etc.).

🧠 Your Task

Train and tune multiple machine learning models and aim for the best F1 Score. For the competition, you will:

• Implement 2 of the following: Random Forest, Support Vector Machine, Neural Network
• Include hyperparameter tuning for improved predictive accuracy
• Conclude with evaluation of the two selected models based on metrics shown in the classification report

🧾 Dataset

• Filename: Students_Grading_Dataset.csv
• Target Column: B_or_Higher

Let's get started!

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from matplotlib import rcParams

rcParams['figure.figsize'] = (15, 8)

Section 1: Preprocessing

Question 1: Load the Academic Performance Dataset

academic_df = pd.read_csv('Academic_Dataset.csv')

Question 2: Show the first 5 rows

academic_df.head()

	Student_ID	Gender	Age	Attendance (%)	Midterm_Score	Final_Score	Assignments_Avg	Quizzes_Avg	Participation_Score	Projects_Score	B_or_Higher	Study_Hours_per_Week	Extracurricular_Activities	Internet_Access_at_Home	Parent_Education_Level	Family_Income_Level	Stress_Level (1-10)	Sleep_Hours_per_Night
0	S1000	0	22	52.29	55.03	57.82	84.22	74.06	3.99	85.90	0	6.2	0	1	1	Medium	5	4.7
1	S1002	1	24	57.19	67.05	93.68	67.70	85.70	5.05	73.79	0	20.7	0	1	3	Low	6	6.2
2	S1003	0	24	95.15	47.79	80.63	66.06	93.51	6.54	92.12	1	24.8	1	1	1	High	3	6.7
3	S1004	0	23	54.18	46.59	78.89	96.85	83.70	5.97	68.42	0	15.4	1	1	1	High	2	7.1
4	S1005	1	21	NaN	78.85	43.53	71.40	52.20	6.38	67.29	1	8.5	1	1	4	High	1	5.0

Question 3: Show the nonnull values for all columns

academic_df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 3206 entries, 0 to 3205
Data columns (total 18 columns):
 #   Column                      Non-Null Count  Dtype  
---  ------                      --------------  -----  
 0   Student_ID                  3206 non-null   object 
 1   Gender                      3206 non-null   int64  
 2   Age                         3206 non-null   int64  
 3   Attendance (%)              2863 non-null   float64
 4   Midterm_Score               3206 non-null   float64
 5   Final_Score                 3206 non-null   float64
 6   Assignments_Avg             2886 non-null   float64
 7   Quizzes_Avg                 3206 non-null   float64
 8   Participation_Score         3206 non-null   float64
 9   Projects_Score              3206 non-null   float64
 10  B_or_Higher                 3206 non-null   int64  
 11  Study_Hours_per_Week        3206 non-null   float64
 12  Extracurricular_Activities  3206 non-null   int64  
 13  Internet_Access_at_Home     3206 non-null   int64  
 14  Parent_Education_Level      3206 non-null   int64  
 15  Family_Income_Level         3206 non-null   object 
 16  Stress_Level (1-10)         3206 non-null   int64  
 17  Sleep_Hours_per_Night       3206 non-null   float64
dtypes: float64(9), int64(7), object(2)
memory usage: 451.0+ KB

Question 4: Show the min, max, median, mean for all columns

academic_df.describe()

	Gender	Age	Attendance (%)	Midterm_Score	Final_Score	Assignments_Avg	Quizzes_Avg	Participation_Score	Projects_Score	B_or_Higher	Study_Hours_per_Week	Extracurricular_Activities	Internet_Access_at_Home	Parent_Education_Level	Stress_Level (1-10)	Sleep_Hours_per_Night
count	3206.000000	3206.000000	2863.000000	3206.000000	3206.000000	2886.000000	3206.000000	3206.000000	3206.000000	3206.000000	3206.000000	3206.000000	3206.000000	3206.000000	3206.000000	3206.000000
mean	0.511853	21.039925	75.391753	70.048958	69.445165	75.186372	74.787645	5.004766	74.853624	0.493450	17.658484	0.301622	0.896132	2.506550	5.459139	6.480162
std	0.499937	1.996322	14.295928	17.089972	17.210758	14.370601	14.601180	2.869489	14.452116	0.500035	7.276542	0.459034	0.305136	1.121811	2.855225	1.457031
min	0.000000	18.000000	50.010000	40.010000	40.000000	50.000000	50.030000	0.000000	50.010000	0.000000	5.000000	0.000000	0.000000	1.000000	1.000000	4.000000
25%	0.000000	19.000000	63.140000	55.442500	54.460000	62.632500	62.250000	2.530000	62.142500	0.000000	11.400000	0.000000	1.000000	2.000000	3.000000	5.200000
50%	1.000000	21.000000	75.730000	70.025000	69.295000	75.255000	74.475000	5.000000	74.930000	0.000000	17.400000	0.000000	1.000000	2.000000	5.000000	6.500000
75%	1.000000	23.000000	87.170000	84.487500	84.102500	87.517500	87.657500	7.530000	87.340000	1.000000	24.100000	1.000000	1.000000	4.000000	8.000000	7.700000
max	1.000000	24.000000	100.000000	99.980000	99.980000	99.980000	99.960000	10.000000	100.000000	1.000000	30.000000	1.000000	1.000000	4.000000	10.000000	9.000000

Question 5: Drop any columns that are not needed

academic_df = academic_df.drop(columns='Student_ID')

Question 6: Address missing values through Imputation or Dropping

academic_df.dropna()
academic_df['Attendance (%)'] = academic_df['Attendance (%)'].fillna(academic_df['Attendance (%)'].mean())
academic_df['Assignments_Avg'] = academic_df['Assignments_Avg'].fillna(academic_df['Assignments_Avg'].mean())

Question 7: Convert the Family_Income_Level column to numeric

family_map = {'Low': 0, 'Medium': 1, 'High': 2}

academic_df['Family_Income_Level'] = academic_df['Family_Income_Level'].map(family_map)

academic_df.head()

	Gender	Age	Attendance (%)	Midterm_Score	Final_Score	Assignments_Avg	Quizzes_Avg	Participation_Score	Projects_Score	B_or_Higher	Study_Hours_per_Week	Extracurricular_Activities	Internet_Access_at_Home	Parent_Education_Level	Family_Income_Level	Stress_Level (1-10)	Sleep_Hours_per_Night
0	0	22	52.290000	55.03	57.82	84.22	74.06	3.99	85.90	0	6.2	0	1	1	1	5	4.7
1	1	24	57.190000	67.05	93.68	67.70	85.70	5.05	73.79	0	20.7	0	1	3	0	6	6.2
2	0	24	95.150000	47.79	80.63	66.06	93.51	6.54	92.12	1	24.8	1	1	1	2	3	6.7
3	0	23	54.180000	46.59	78.89	96.85	83.70	5.97	68.42	0	15.4	1	1	1	2	2	7.1
4	1	21	75.391753	78.85	43.53	71.40	52.20	6.38	67.29	1	8.5	1	1	4	2	1	5.0

Section 2: Data Exploration

Question 1: Create a pairplot of the dataset

sns.pairplot(academic_df, hue='B_or_Higher', palette='husl')
plt.show()

Question 2 [Text]: Make 2 observations from the pairplot

• Student with parent education level 3 are more likely to have B or higher grade.
• There are moderately more number of students achieved B or higher with no internet access than with it.

Question 3: Create a heatmap of the variables (hint: pass the argument "fmt='.3f'" into the heatmap for improved readability)

sns.heatmap(academic_df.corr(), annot=True, cmap='YlGnBu', fmt='.3f')
plt.show()

Question 4 [Text]: What do you notice from the heatmap?

• There's a moderately strong correlation of 0.605 between Attendance and B or higher.
• There's little correlations between all other variables.
• No multi collinearlity occurs.

Section 3: Model Creation

Question 1: Create an 'X' variable and the target 'y' variable ("B_or_Higher")

y = academic_df['B_or_Higher']
X = academic_df.drop(columns='B_or_Higher')

Question 2: Do a train test split of the dataset with a test size of 30%

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)

Question 3: Scale the training and testing data

from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()

scaler.fit(X_train)

X_train_scaled = scaler.transform(X_train)
X_train_scaled = pd.DataFrame(X_train_scaled, columns=X_train.columns)

X_test_scaled = scaler.transform(X_test)
X_test_scaled = pd.DataFrame(X_test_scaled, columns=X_test.columns)

Question 4: Select and fit a model for the dataset, then show accuracy

X_train_scaled.isnull()

	Gender	Age	Attendance (%)	Midterm_Score	Final_Score	Assignments_Avg	Quizzes_Avg	Participation_Score	Projects_Score	Study_Hours_per_Week	Extracurricular_Activities	Internet_Access_at_Home	Parent_Education_Level	Family_Income_Level	Stress_Level (1-10)	Sleep_Hours_per_Night
0	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False
1	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False
2	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False
3	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False
4	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
2239	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False
2240	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False
2241	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False
2242	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False
2243	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False	False

2244 rows × 16 columns

from sklearn.metrics import accuracy_score
from sklearn.linear_model import LogisticRegression

logmodel = LogisticRegression(max_iter=1000)
logmodel.fit(X_train_scaled,y_train)
log_preds = logmodel.predict(X_test_scaled)

print(accuracy_score(y_test, log_preds))

0.7463617463617463

Question 5: Select and fit a second model for the dataset, then show accuracy

from sklearn.ensemble import RandomForestClassifier

rfc = RandomForestClassifier(n_estimators=100)

rfc.fit(X_train_scaled, y_train)
rfc_preds = rfc.predict(X_test_scaled)

print(accuracy_score(y_test, rfc_preds))

0.7806652806652806

Section 4: Model Evaluation

Question 1: Show the confusion matrix and classification report for the first model

from sklearn.metrics import classification_report, confusion_matrix

cm = confusion_matrix(y_test, log_preds)
print(cm)

print(classification_report(y_test, log_preds))

[[378 134]
 [110 340]]
              precision    recall  f1-score   support

           0       0.77      0.74      0.76       512
           1       0.72      0.76      0.74       450

    accuracy                           0.75       962
   macro avg       0.75      0.75      0.75       962
weighted avg       0.75      0.75      0.75       962

Question 2: Show the confusion matrix and classification report for the second model

rcm = confusion_matrix(y_test, rfc_preds)
print(rcm)

print(classification_report(y_test, rfc_preds))

[[473  39]
 [172 278]]
              precision    recall  f1-score   support

           0       0.73      0.92      0.82       512
           1       0.88      0.62      0.72       450

    accuracy                           0.78       962
   macro avg       0.81      0.77      0.77       962
weighted avg       0.80      0.78      0.77       962

Question 3 [Text]: Which model performed best of the two?

The Random Forest Classifier performed the best compared to the Logistic Regression in terms of precision, recall and accuracy (with an accuracy of 0.78 vs 0.75).

Question 4 [Text]: Evaluate the performance of your best model and explain what you can conclude from the precision and recall

The Random Forest Classifier had a wider range in terms of precision and recall values for the different classes: It had significantly higher precision (True Positives/Predicted Positives) for class 1 of students who earned a B or Higher, and had a bit higher recall (True Positives/Total Positives) for class 0 of students who did not earn a B or Higher. As a result, the model's performance would be more erratic if there was a larger class imbalance.

Question 5: Compute and compare the Adjusted Rand Index (ARI) for both models

from sklearn.metrics.cluster import adjusted_rand_score

print("Logistic ARI: ",adjusted_rand_score(y_test, log_preds))
print("RFC ARI: ",adjusted_rand_score(y_test, rfc_preds))

Logistic ARI:  0.24199044447051332
RFC ARI:  0.3141327008121759

Question 6 : Why is comparing different models statistically complex, and how does ARI help?

• Models take different assumptions about the data, such as linearity and normality. They can be more complex when adding more variables, leading to overfitting. It also can be hard to interpret as models can have different metrics.

• ARI help by validates clustering results, identify the most effective algorithms and parameters, avoid misleading interpretations of clustering results

Question 7 [Text]: Discuss how normalization of results (such as scaling or standardizing metrics) allows for fair model comparison and why this is important

It ensures that all features contribute equally to the analysis. It's important because preventing larger-scale features from dominating smaller ones is crucial for an unbiased results and improved model performance.